nlesc-jcer / quicklb Goto Github PK

Create an artificial example demonstrating the load-balancing benefits. We envision an integration time step of a grid (1-D or 2-D) consisting of two stages during the time stepping:

1. A 'background' time step process that has uniform performance across the mesh, representing the transport calculations in CFD.
2. A 'signal' process that is more random and unpredictable across the mesh, representing chemistry and thermodynamics that should be balanced

These different performance signatures can be achieved by halting the process for particular amount of time, or some other way (e.g. doing a useless numerical calculation). We use a simple Cartesian MPI parallel setup for the toy model representing the mesh partitioning based on fluid transport balancing.

Create a documented program that runs this toy model and a program that balances the model using quicklb, showcasing the capabilities of the library.

Python

• loadbalancer class with numpy arrays
• direct interface with numpy arrays
• Example with offloaded and non-offloaded work together

Fortran

• any example

Describe dependencies and install instructions on the Github page. Add a more elaborate API description on using readthedocs or similar. Also the examples should be well-documented.

Also we should contribution guidelines, citation.cff and other stuff listed in the NLeSC guide.

Translate the example of issue #1 to python